Messages in this thread |  | | From | (Eric W. Biederman) | Date | Tue, 08 Jan 2013 16:58:14 -0800 | Subject | Re: [PATCH v7u1 26/31] x86: Don't enable swiotlb if there is not enough ram for it |
| |
Konrad Rzeszutek Wilk <konrad.wilk@oracle.com> writes:
> On Tue, Jan 08, 2013 at 03:40:11PM -0800, Yinghai Lu wrote: >> On Mon, Jan 7, 2013 at 7:50 PM, Yinghai Lu <yinghai@kernel.org> wrote: >> > On Mon, Jan 7, 2013 at 7:13 PM, Eric W. Biederman <ebiederm@xmission.com> wrote: >> >> I meant we should detect failure to allocate bounce buffers in in >> >> swiotlb_init() instead of panicing. >> >> >> >> I meant swiotlb_map_single() should either panic or simply fail. >> >> >> >> If I have read lib/swiotlb.c correctly the only place we allocate a >> >> bounce buffer is in swiotlb_map_single. If there are more places we can >> >> allocate bounce buffers those need to be handled as well. >> > >> > ok, will give it a try. >> >> please check if you are ok with attached. >> >> looks like it need more change of lines. > > The swiotlb_full check I don't believe is neccessary. You won't ever get > to that unless swiotlb_map_page has at least provided a bounce buffer. > And if the swiotlb_map_page does not have a bounce buffer it will exit > with: > > + if (no_iotlb_memory) > + return SWIOTLB_MAP_ERROR; > + > > which is dangerous. That is b/c there are drivers that don't use the > dma_mapping_error check (so check the bus address after calling > pci_map_*). This means they would try to do DMA on 0xffffffff (yikes!). > > That is reason the failback (v_overflow_buffer) is still in > usage - b/c we have drivers that might just do this and this is the last > resort for them. And until those drivers are fixed - we _need_ this > fallback to work.
So instead we need to say?
+ if (no_iotlb_memory) + panic("Cannot allocate SWIOTLB buffer"); +
Which is just making the panic a little later than it used to be and seems completely reasonable.
Eric
|  |